Date of Award

2017

Document Type

Open Access Master's Thesis

Degree Name

Master of Science in Forestry (MS)

Administrative Home Department

College of Forest Resources and Environmental Science

Advisor 1

Robert E. Froese

Committee Member 1

Nan Pond

Committee Member 2

Curtis Edson

Abstract

Active remote sensing technology (LiDAR) and passive remote sensing technology (Pleiades and Göktürk-2 satellites) were used to find a meaningful relationship between ground data and remote sensing instruments for Istanbul Forest, Turkey. Two dominant species in the field, oak (deciduous trees) and maritime pine (coniferous trees), were researched. There were 86 plots total, 41 for maritime pine and 45 for oak. Three diameter at breast height (DBH) thresholds were studied. Trees of any DBH (DBH≥0.1 cm), trees ≥8 cm DBH thresholds and, trees ≥10 cm DBH thresholds. Both satellite image metrics were derived from Grey Level Co-occurrence Measures (GLCM). All metrics derived from satellite images and LiDAR data were incorporated into a hybrid approach. All metrics were separated and compared to each other to investigate how they are functioning separately. Linear regression, randomForest, and randomForest imputation models were used. The best R2 was 0.90 using three remote sensing instruments for tree counts based on the plot level for oak species. The highest % explained variances were 67.15% for tree count based on the plot level for oak species in randomForest model and 55.85% for tree count based on the plot level for oak species in randomForest Imputation. LiDAR data had a better relationship with ground data. Band 2 and band 4 of both satellite images were stronger predictors for deciduous trees. Band 3 and band 4 of both satellite images were used more for coniferous trees. Some of the most useful GLCM options were entropy for deciduous trees and correlation, variance and second moment for coniferous trees.

Download

Share

COinS